Tools
Tools
Lohano, Jackay (2024) Automated Detection of Fake News in Urdu Language Using Pre-Trained Transformer Models. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (764kB) | Preview |
Abstract
The propagation of misinformation across different languages and domains on various social media platforms is of grave concern for societies and individuals due to its wide-range consequences. Although, researchers have addressed this challenge using advanced deep learning (DL) models, fake news detection in low resource languages such as Urdu is still at nascent stage. Studies have used traditional machine learning (ML) models on a very small and domain-restricted Urdu datasets for fake news detection. This study explores fake news classification in Urdu using three state-of-the-art (SOTA) pre-trained multilingual transformer modelsi.e. mBERT, DistilmBERT and mT5 on a large and multidomain Urdu dataset. Models are evaluated on the evaluation metrics such as accuracy, precision, recall and f1-score. The results show that DistilmBERT model demonstrates promising results with an accuracy of 89% compared to its larger counterpart mBERT and mT5 models. The findings reveal the potential of DistilmBERT model for real-world applications in memory-constrained environments for identification of fake news. This research contributes to the ongoing efforts to combat misinformation in resource-scarce languages by building a reliable Urdu fake news detection model, addressing the complexities of information dissemination and manipulation in modern age.
Actions (login required)
 |
View Item |